In a nutshell
- The YouTube algorithm prioritizes valued watch time, rewarding videos that keep viewers engaged and satisfied.
- Key metrics like click-through rate and audience retention are crucial for creators, as these factors determine how well a video performs in the YouTube algorithm and its subsequent reach.
- For new creators, focusing on a specific niche and consistent content creation is more beneficial than striving for perfection in each video, as this helps the algorithm understand and target the right audience for their content.
To widen their chances of receiving attention on the platform, creators must work with the YouTube algorithm. That is the simple reality of YouTube. However, there is a certain mystery surrounding how the algorithm actually works.
Let’s look into how the YouTube algorithm decides whether your latest video will reach the eyes and ears of thousands of potential fans, subscribers and customers, or be doomed to gather dust on the digital shelf along with millions of others. We’ll start at a high level, analyzing the primary goals of the YouTube algorithm. Then we’ll zoom into the finer details of the metrics and mechanics behind it, so you can make the algorithm work for you to take advantage of the potentially unlimited reach that YouTube can offer.
The main goal of the YouTube algorithm
YouTube is a business that runs primarily on advertising revenue. They have a strong financial incentive to keep viewers watching for as long as possible. Like that friend who insists you take another slice of cake, even though you’ve already had three. It seems like the algorithm is designed so that you click, watch, click, watch and repeat until you forget what sunlight feels like.
Now, YouTube Public Relations would beg to differ (slightly). In an article by YouTube’s VP of Engineering and Recommendations Cristos Goodrow, he explains how the algorithm is not designed just to maximize watch time. Rather, they have a concept called “valued watch time” that is used as the primary goal and North Star metric for their recommendation system, more widely known as the YouTube algorithm.
“We don’t want viewers regretting the videos they spend time watching and realized we needed to do even more to measure how much value you get from your time on YouTube,” said Goodrow.
Valued watch time is calculated from pure watch time using data extrapolated from user surveys. If you’ve spent enough time watching YouTube, you’ve probably seen one.
The surveys don’t come up that often, but apparently, YouTube receives millions of survey results daily (don’t forget they serve billions of videos per day.)
It makes sense for YouTube to want to make its viewers happy. Happy viewers come back to watch more videos (and ads) or even sign up for a premium subscription. So, if your videos can make your viewers happy, whether through entertainment, education or giving them whatever it is they were searching for when they came across your video, the algorithm will reward you accordingly.
How does the algorithm decide which videos to recommend?
Recommending the most popular videos
This is the most unsophisticated and brute-force technique when it comes to recommendations: “Just show them what everybody else watched.” But it undoubtedly still plays a large part in the highly sophisticated algorithm used to this day.
To see what we mean, just try logging on to YouTube from an Incognito or Private browser window. We guarantee you’ll be offered at least one MrBeast video, probably one with over a hundred million views. There will probably be some Selena Gomez or other popstar clips, again with millions of views, and perhaps some current sporting events that have been attracting the attention of the masses within the last few hours or days. Without any data or browsing history, it makes sense for YouTube to fall back on recommending what’s popular or trending.
YouTube’s fallback strategy: Recommend the most popular videos
Popularity can also be a proxy for authority. Imagine you are searching on YouTube for a way to fix your leaking toilet. You’re presented with two video choices, both claiming to help you solve your problem. One has 4.4 million views, and the other has 87 views. Which one would you be more likely to click? Almost assuredly the 4.4 million one.
That might seem like an extreme example, and you might wonder why we’re stressing the point that popularity is an advantage when you can’t just flick a switch and become popular overnight. How does it help you to know that a beginning YouTube channel competing for views is like a tiny firefly dancing for attention beneath the Times Square fireworks display?!
Well, to start with, it helps to set realistic expectations. It helps to know that if your first few videos only get a handful of views it doesn’t mean you failed or that they’re not good videos.
Also, it helps to realize that you need to pick a small target audience and focus when starting out. You can’t compete with MrBeast, so don’t try. You probably can’t even compete with the 4.4 million view toilet fixer. But if there’s a specific model of toilet that you know better than anyone else, then focus on helping people with that specific toilet (you can substitute “toilet” with whatever is relevant to your channel), and again, YouTube will reward you for making its viewers happy.
Using clustering techniques to match videos to users
Moving beyond the most basic strategy of recommending popular videos, the algorithm uses clustering to group viewers and videos. A clustering algorithm is any algorithm that groups objects into clusters that are in some sense “similar.”
As an example, below is a cluster analysis I, the author of this article, performed on articles written on medium.com. Each bubble represents a “tag” (like a hashtag), and bubbles are arranged into clusters based on which tags were used together.
The primary metric the YouTube algorithm uses is consecutive viewing, or watch history. If two videos are regularly watched together, then they’re clustered together, and viewers of one video are likely to recommend the other. Similarly, two viewers who watched and enjoyed the same video, or set of videos, are likely to be recommended similar videos.
Again, it’s a pretty simple concept but surprisingly powerful. With enough data (and YouTube gets literally billions of video views per day), YouTube can quickly learn which viewers will likely enjoy which videos. For those viewers that are logged in to YouTube with their Google account, YouTube has access to a rich history of every video they clicked on, scrolled past, watched, liked, disliked, shared, etc., so it can predict what you are likely to want to watch before you even think about typing into the search bar.
What does this mean for you as a video creator?
It means that you need to think carefully about which viewers have watched and enjoyed your previous videos. These people include your subscribers but are not limited to your subscribers, especially if you only have a few hundred. The people who have watched your most recent videos are the first people who will recommend your new video. If these people choose to click, watch and like your new video, then it will be distributed more widely. If they don’t, it won’t.
Judging the quality of each video
There are several metrics YouTube uses to decide whether viewers receive a video well. The two most important metrics are click-through rate (CTR) and audience retention.
Click-through rate (CTR)
There is a good reason why this was the number one ranking factor mentioned by the YouTube VP of Engineering. CTR is calculated as the percentage of people who click on your video link once it is presented in their feed or search results. The YouTube recommendation panel is a highly valuable screen for real estate. The YouTube algorithm aims to fill it with videos with a strong probability of being clicked to keep viewers on the platform and engaged.
YouTube provides creators with each video’s CTR in the YouTube Studio app under the Reach tab and tells you that these numbers need to be high for the algorithm to work.
How high is high enough for a CTR? That depends on your niche, experience, and the size of your channel. Google’s YouTube Help page tells us, “Half of all channels and videos on YouTube have an impressions CTR that can range between 2% and 10%.” So, a CTR above 10% is great, while a CTR less than 2% probably means your video will not continue to be widely distributed. Once you’ve uploaded five or 10 videos, you will begin to see what an average CTR looks like for your channel. Then, aim to beat that average on your next video.
How to improve your CTR? It’s counterintuitive, but it really does not have anything at all to do with the actual quality of your video. After all, when the viewer decides to click — or not — they haven’t started watching it yet. So, the key triggers you have to pull are the title and the thumbnail.
The title and thumbnail should entice a potential viewer to click by taking advantage of their curiosity and interest or generating an emotional reaction. You’ll notice that most successful thumbnails contain a face and that face conveys emotion (excitement, curiosity, fear, etc.). Look at the titles and thumbnails of the top-ranking videos in your niche, and you will start to see what works. What do your viewers want? What motivates them? What excites them?
The best titles and thumbnails seem to border on “clickbait,” and you need to be careful not to cross that line, but it can help to go close. After all, you really do want them to click. As a general rule, if your video offers something in the title, then you should deliver it in the video. Ideally, make it clear that you will deliver it within the first 30 seconds. If the viewer clicks in but then clicks away within a few seconds, your audience retention statistics will be poor, and a high CTR but low retention won’t be enough to make the algorithm happy.
The second key metric is audience retention. Basically, this measures, for those people who did choose to click on your video, how much of it did they actually watch?
YouTube Studio gives you a retention chart like the one shown below for each video you publish. In the example below, 40% of viewers watched after 2 minutes and 29 seconds.
It’s normal to lose up to 50% of viewers within the first 30 seconds. People are busy, and people have short attention spans. Again, think of how you interact with YouTube as a viewer. Do you usually watch all of every video you click? Or, more likely, do you scroll, click, scroll, click, watch, like, scroll, click?
This is why it’s so important to start your video strong. Get to the point; don’t waste words. If you can keep 70% of viewers for the first 30 seconds, and 50% of the viewers until the end, the algorithm will love it and reward you for it.
Again, this emphasizes the need to avoid clickbait in your title. Don’t promise something that you cannot deliver. If somebody clicks in and only watches three seconds, that is a strong negative signal to the algorithm, so it would actually have been better if that person had not clicked in the first place. This is why you should be careful about posting your YouTube video on Twitter, Reddit or another social platform where users are quickly scrolling through lots of three-second posts. How likely are they to stop and watch your video for several minutes?
So, if your CTR is high and your audience retention is high, your video will be distributed well by the algorithm. Again, what “high” means is relative. Let’s say that if your CTR is higher than average and your audience retention is better than average for your channel, that video will be distributed more widely compared to others on your channel.
YouTube Shorts may have started out as a TikTok competitor, but now they are undeniably popular in their own right. You can’t avoid Shorts now on YouTube. Not only are they heavily recommended on the YouTube home page, Shorts can even show up in response to regular Google searches.
The algorithm for YouTube Shorts is similar to that described above, but has some differences that are worth mentioning. First, CTR is not so relevant here because most viewers don’t get to see a Short by clicking on a thumbnail and title. Rather, they just get it served to them as part of a feed when they swipe up from the previous Short. Audience retention is still critical, but the time frame is different. Rather than looking at the first 30 seconds, it’s the first 2 seconds that matter. You need to stop the user from swiping. Most viewers swipe through several Shorts before actually stopping to watch one.
Something interesting needs to happen within the first 2 seconds. Begin with an attention-grabbing introduction that immediately sparks interest. Utilize visual cues, text overlays and dynamic transitions to retain attention. Delivering value in a concise format is key, so focus on delivering a clear message or entertaining moment that resonates with your target audience. If your Shorts have a narrative or educational component, tease the content’s value upfront and promise a satisfying conclusion. Strong retention rates signal to the algorithm that your content is engaging, increasing the likelihood of wider distribution.
Beyond that, if you can involve viewers by encouraging likes, comments and shares, this also helps the algorithm promote your Short. Engaging with your audience through comments can trigger a positive feedback loop, signaling to the algorithm that your content is resonating with viewers and warranting wider distribution.
One thing to remember with Shorts is that the demographic of Shorts viewers may differ from your regular target audience. By differ, we primarily mean younger. But this is likely to change as short-form content becomes increasingly popular with all demographics.
Tips for starting out on YouTube
The advice given above about focusing on your niche and pleasing your target audience is harder to implement if you have a very small channel. This is because YouTube just doesn’t have enough data on your audience to extrapolate to other people who may like your content. If the algorithm shows your video to 100 people and they happen to be the wrong people for your video, it may not get many more views beyond this.
So, when starting out, consistency and quantity are more important than actual quality. Of course, your videos still need to be good. But, for example, if your video is 90% in terms of quality and you are considering spending another four hours to get the quality to 100%, you may be better off publishing the 90% video and using that time to create more content.
Once YouTube gets more data on your audience, the algorithm will be able to serve you better and show your videos to more of the right people, and that’s when the quality really matters.
The YouTube algorithm is designed primarily to serve viewers, not creators, but at the same time, there are adjustments in the algorithm designed to help identify the right target audience for small channels. YouTube assures that:
“We do care about small creators. We actually have teams that are dedicated to making sure small creators can still break through on YouTube.” Todd Beaupré said at VidCon 2022
So, don’t be disheartened if your first few — even your first 30 — videos don’t get as many views as you think they deserve. Focus on consistency, and you should start to see the views and audience retention improving over time.
How to make the algorithm work for you
In summary, aim to continually improve your CTR, audience retention and viewer satisfaction. The key to success lies in delivering value to viewers by identifying a target audience and niche, becoming a trusted guide in that domain, and consistently offering satisfying content that aligns with viewers’ interests.
You may have heard the phrase: “You’re just one video away.” This saying means that one video can be all it takes to shine the spotlight on your channel and attract the audience you are aiming for. That’s the magic and power of YouTube. They have such a large viewer base that once one video goes viral and grows exponentially, there is really no limit to the number of people your content can reach.
Good luck, and enjoy the journey.
Contributing authors to this article include Russel Lim and Jo Mayer.
Image courtesy: Deep Neural Networks for YouTube Recommendations